Mediation Analysis with Multiple Mediators.

Area of Science:

• Causal inference
• Biostatistics
• Epidemiology

Background:

• Traditional mediation analysis often assumes single mediators and linear relationships.
• Recent advances address interactions and non-linearities but may not cover multiple mediators.
• Estimating effects through multiple interacting mediators remains a challenge in causal inference.

Purpose of the Study:

• To extend causal inference methods for mediation analysis to settings with multiple mediators.
• To propose and evaluate analytic approaches for estimating effects mediated through multiple pathways.
• To develop methods robust to exposure-mediator and mediator-mediator interactions.

Main Methods:

• Developed two analytic approaches: one based on regression and one based on weighting.
• The methods accommodate exposure-mediator interactions and, to some extent, mediator-mediator interactions.
• Handled binary or continuous mediators and binary, continuous, or count outcomes.

Main Results:

• Proposed methods can estimate effects mediated through multiple mediators and other pathways.
• The approach remains valid even when mediators affect one another, unlike single-mediator strategies.
• Provided a characterization for when the sum of separate mediated effects equals the joint mediated effect.

Conclusions:

• The proposed causal inference framework effectively handles multiple mediators in mediation analysis.
• The methods are robust to unmeasured common causes of multiple mediators.
• Offers a robust approach for complex mediation scenarios in various research fields.